Apple vs. ChatGPT, Claude, and Gemini: why Apple isn't playing the same game

The image sums it up better than any explanation: inside Xcode, the tool used to build Apple apps, the developer chooses which artificial intelligence to work with. The options are Claude, from Anthropic; Codex, from OpenAI; and Gemini, from Google. Apple did not build a rival to dethrone them. It put all three in a list for you to choose from. That decision explains why Apple is not playing the same game as the rest of the industry.

Three different bets

In recent years, every major technology company reached the same conclusion: having its own frontier artificial intelligence model was a matter of survival. OpenAI built GPT. Anthropic, Claude. Google, Gemini. Meta, Llama. Microsoft built its own family. For all of them, the model is the core asset, the moat that protects them, and they compete over the same things: the best reasoning, the longest context, the lowest price.

Apple chose the opposite path. Instead of building its own frontier model, it signed a deal with Google, reported by Bloomberg at around one billion dollars per year, to use a custom version of Gemini that it then distills into a smaller model capable of running on the device. How that architecture works internally is a separate story, but the strategic headline is clear: the intelligence powering the new Siri is not Apple's.

Distribution as a moat

If Apple is not competing on the model, the question is what it is competing for. The answer is the place where the model runs. The underlying bet is that the model layer is becoming a commodity: GPT, Claude, and Gemini are increasingly similar in what they can do, and the distance between them shrinks with every version. If they end up being interchangeable, whoever controls the gateway through which the user reaches them captures the value.

It is Apple's classic playbook: do not make the content, own the store. And for that, it has two things no AI lab can buy: around two billion active devices and the user's personal context — emails, photos, messages, and files — accumulated over years and processed with privacy as the banner, partly on the device and partly on its own servers.

That is why the new Siri does not even try to be a conversational chatbot. It is brief, gets to the point, and avoids the extra chatter that characterizes ChatGPT. It is not fighting to be the best conversationalist; it is fighting to be the one that gets things done on your phone. And in iOS 27 it goes one step further: rival assistants can plug into Siri, and the user can choose Claude or Gemini for tasks like writing or generating images. Apple does not want to be the model. It wants to be the place where you choose the model.

The risk of renting the brain

The flip side of this reading is uncomfortable. Depending on Google for the part it could not build on its own is not a strength; it is a vulnerability. Apple arrived late and gradually, with cautious steps, while the rest of the industry has already made familiar what Apple is only now showing. At WWDC 2026, there was no moment that suddenly put it ahead of OpenAI, Google, or Anthropic.

The question, then, is whether this is a strategy or a resignation. Is Apple the smart owner of the place where everything runs, or did it end up renting the brain from Google because it had no other choice?

Both things can be true at the same time. Apple lost the race for the best model and, instead of continuing to run from behind, switched tracks: it is betting that distribution and integration are worth more than raw power. If models become commodities, it is the smartest move of the decade. If capability continues to make the difference, Apple is left as the intermediary reselling other companies' intelligence. The Xcode selector, with three rivals turned into options on a list, is the exact image of that bet: Apple does not want to win the race; it wants to be the track where everyone runs.